Atlas / MCP servers / buzzni / claude-memory-layer

claude-memory-layerBLOCK

Name: claude-memory-layer
Author: buzzni

mcp/buzzni/claude-memory-layer

MCP server by buzzni · Audited 2026-10-12

⚑ Report / re-scan Source ↗

Verdict

BLOCK

Grade

Trust score

48 /100

Exposed tools

48 31r · 15w · 2d

Transport

stdio

License

—

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

Claude Code, Codex, Hermes, MCP 클라이언트가 함께 쓰는 프로젝트 스코프 메모리 레이어입니다. 대화/세션 기록을 SQLite 이벤트 저장소와 LanceDB 벡터 인덱스로 관리하고, 작업 시작 시 필요한 기억·타임라인·액션 프런티어를 privacy-safe 컨텍스트로 제공합니다.

개요

Claude Memory Layer는 AI 에이전트의 대화와 작업 이벤트를 프로젝트별로 저장하고, 새로운 작업을 시작할 때 관련 기억을 자동 검색해 컨텍스트로 제공합니다. 최신 npm 패키지는 단순 Claude Code hook을 넘어 다음 용도로 사용할 수 있습니다:

연속성 있는 대화: 이전 세션에서 논의한 결정, 오류, 검증 결과를 검색
프로젝트 맥락 이해: 프로젝트별 isolated store와 projectPath 기반 MCP/CLI 조회
다중 에이전트 호환: Claude Code hooks + Codex/Hermes session import + MCP stdio server
운영형 메모리: actions/frontier/checkpoints/facets/lessons/retention audit로 다음 작업을 이어받기
Perspective Memory: 사용자/어시스턴트/서브에이전트 관점의 actor card와 observation 관리
신뢰 가능한 벡터 동기화: Vector Outbox V2, worker lock, dashboard/vector-status health로 누락 복구

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add claude-memory-layer --env ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY} --env DART_API_KEY=${DART_API_KEY} --env FINNHUB_API_KEY=${FINNHUB_API_KEY} --env FRED_API_KEY=${FRED_API_KEY} -- npx -y [email protected]

claude-desktop

{
  "mcpServers": {
    "claude-memory-layer": {
      "command": "npx",
      "args": [
        "-y",
        "[email protected]"
      ],
      "env": {
        "ANTHROPIC_API_KEY": "${ANTHROPIC_API_KEY}",
        "DART_API_KEY": "${DART_API_KEY}",
        "FINNHUB_API_KEY": "${FINNHUB_API_KEY}",
        "FRED_API_KEY": "${FRED_API_KEY}"
      }
    }
  }
}

Exposed tools (48)

31 read · 15 write · 2 destructive. Blast radius: 2 tools can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.

Tool	Risk	Description
`external-market-context`	read	Read-only external market/company context from DART, FRED, and Finnhub with structured MarketContextSnapshot bull/bear/risk/catalyst analysis.
`golden-memory-usefulness-v1`	read	privacy-safe fixture
`longmemeval-turn-retrieval-smoke-hybrid-source`	read	Turn-level LongMemEval fixture used internally for hybrid session retrieval.
`mem-action-list`	read	List project-scoped memory actions with compact status/frontier metadata and no raw transcript disclosure.
`mem-action-update`	write	Update a project-scoped memory action status through an audited, state-changing operation.
`mem-actor-card-get`	read	Read the compact actor card for one observer -> observed perspective without raw transcript disclosure.
`mem-actor-card-upsert`	write	Create or replace a compact actor card through an audited state-changing perspective operation.
`mem-actor-list`	read	List project-scoped or global memory actors with compact privacy-safe identity metadata.
`mem-asset-bind`	write	Create, update, or disable an actor-to-asset injection binding after a bind permission check.
`mem-asset-catalog-sync`	write	Preview or apply idempotent registration of canonical project lessons and core-memory blocks as private memory assets. Preview is the default and never changes canonical content.
`mem-asset-check`	read	Explain whether the requester has a specific memory asset permission and which policy rule decided it.
`mem-asset-create`	write	Register a project-scoped memory asset. The requester becomes its owner; content stays in the referenced canonical subsystem.
`mem-asset-get`	read	Get one memory asset only when the requester has read access; unauthorized and missing assets return the same not-found shape.
`mem-asset-grant-set`	write	Replace one actor\
`mem-asset-list`	read	List only memory assets readable by the requester; inaccessible private assets are omitted.
`mem-asset-update`	write	Update memory asset metadata after an owner or explicit write-grant check. Supports optimistic version checks.
`mem-checkpoint-create`	write	Create an audited project-scoped checkpoint for a resumable operation target without exposing raw local paths or transcript content.
`mem-checkpoint-list`	read	List project-scoped checkpoints with bounded metadata for safe task resumption.
`mem-context-pack`	read	Build a compact, agent-ready project context pack from relevant memories plus recent project timeline. Use at the start of Hermes/Codex work to recover the important project state without reading raw transcripts.
`mem-core-block-get`	read	Read this project\
`mem-core-block-update`	write	Replace one core memory block through an audited write. Registered blocks require write access when server enforcement is enabled.
`mem-details`	read	Get full content of specific memories. Use after mem-search to get complete information. Conditional mutation: each successfully resolved reference records privacy-safe navigation telemetry.
`mem-entity-supersede`	read	Mark one entity (task/condition/artifact/source_file) as superseded by another through an audited fact-reconciliation write. The old entity stops being traversed/surfaced by graph-based retrieval (mem-graph-query and related retrieval lanes); it is not deleted, just excluded from
`mem-facet-query`	read	Query project-scoped memory facets and return compact privacy-safe facet rows for filtering or inspection.
`mem-facet-tag`	write	Attach or update a project-scoped facet tag on a memory operations target with governance audit metadata.
`mem-frontier`	read	Return the project-scoped execution frontier: next actions, blocked work, checkpoints, and safe resume hints.
`mem-graph-query`	write	Run bounded project-scoped graph expansion/query diagnostics over active entities and edges with sanitized explanations.
`mem-import-latest`	write	Explicitly import the latest local Claude Code, Codex, and/or Hermes session history into project-scoped memory before retrieving context. This mutates memory; use for freshness jobs before mem-context-pack.
`mem-lesson-candidates`	read	Detect not-yet-saved reusable workflow patterns (rule-based, from repeated successful tool sequences across sessions) so they can be reviewed and promoted with mem-lesson-save. Read-only — does not write a lesson.
`mem-lesson-get`	read	Fetch one curated project lesson by lessonId or project-unique name — the body behind an entry in the session-start lesson index. Registered assets are permission-filtered when server enforcement is enabled.
`mem-lesson-list`	read	List project-scoped procedural lessons as compact skill/runbook candidates. Registered assets are permission-filtered when server enforcement is enabled.
`mem-lesson-save`	write	Save a reviewed project-scoped lesson or rule as a curated memory. Updates to registered lessons require write access when server enforcement is enabled.
`mem-perspective-context`	read	Build a compact actor-card plus perspective-observation context bundle for one observer -> target actor pair.
`mem-perspective-observation-create`	write	Create an audited observer -> observed perspective observation with bounded evidence references.
`mem-perspective-observation-delete`	destructive	Soft-delete an audited perspective observation by id; hard deletion is not exposed.
`mem-perspective-query`	read	Query observer -> observed perspective observations as a separate retrieval lane with source-reference hints.
`mem-project-timeline`	read	Summarize recent project memory by session, source agent, event counts, and safe previews. Useful for understanding what happened recently before continuing work.
`mem-retention-audit`	write	Run a dry-run retention governance audit for project-scoped memories and return policy explanations without hard deletion.
`mem-search`	read	Search claude-memory-layer for relevant past conversations and insights. Returns a compact index of results - use mem-details to get full content.
`mem-shared-actor-link`	read	Explicitly link this project-local actor to a shared principal before using actor-scoped cross-project troubleshooting search. This is a global shared-store mutation; it does not grant canonical memory-asset permissions.
`mem-shared-actor-status`	read	Inspect whether this project-local actor has an explicit shared principal link. No shared troubleshooting content is returned.
`mem-shared-actor-unlink`	destructive	Remove this project-local actor\
`mem-shared-asset-get`	read	Read one active canonical lesson or core-memory block from another project only after both named actors are explicitly linked to the same shared principal. The source asset must be registered, active, visibility=shared, and readable under the source project policy. This never auto-injects content.
`mem-shared-search`	read	Search verified shared troubleshooting entries only from projects explicitly linked to requesterActorId through the same shared principal. An unlinked actor receives an empty result.
`mem-stats`	read	Read-only: get statistics about an existing memory store without creating or migrating it.
`mem-timeline`	read	Get chronological context around specific memories. Useful for understanding the conversation flow.
`private-real-session-qrels`	read	contains raw real session text that reports must not leak
`retrieval-review-golden-promotion-candidates`	read	Privacy-safe candidate plan for manually promoting bad retrieval review queue items into the golden replay fixture.

Trust audit

BLOCKgrade F · trust 48/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	WARN
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	PASS
L3	Class-specific surface	WARN
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (11 observation(s))
Network: declared (7 observation(s))
Shell: declared (7 observation(s))
Dependencies: not all pinned
Secrets in source: found

Findings (25)

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

scripts/evaluate-lesson-hybrid-heldout.ts:34

const dynamicImport = new Function('specifier', 'return import(specifier)') as (specifier: string) => Promise<unknown>;

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

scripts/evaluate-lesson-hybrid.ts:19

const dynamicImport = new Function('specifier', 'return import(specifier)') as (specifier: string) => Promise<unknown>;

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

src/apps/cli/index.ts:3950

exec(command, (error) => {

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

src/extensions/vector/embedder.ts:411

const dynamicImport = new Function('specifier', 'return import(specifier)') as (

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

MEDIUMInventory / provenance · inv.binary · CWE-1104

.DS_Store

.DS_Store

Why it matters. a compiled or binary member cannot be reviewed from source

Fix. ship source, or explain the binary in the README

MEDIUMInformation disclosure · disclose.log_secret · CWE-209, CWE-532

src/apps/cli/index.ts:3911

console.log(`🔐 Password: ${password ? 'enabled' : 'disabled'}`);

MEDIUMNetwork egress · net.raw_ip · CWE-200, CWE-319

scripts/dashboard-smoke.ts:63

const baseUrl = `http://127.0.0.1:${port}`;

MEDIUMNetwork egress · net.raw_ip · CWE-200, CWE-319

src/apps/server/index.ts:425

const response = await fetch(`http://127.0.0.1:${port}/health`);

MEDIUMHard-coded secrets · secret.anthropic · CWE-798, CWE-321

tests/core/privacy-filter.test.ts:111

['anthropic', 'sk-ant-ABCDEFGHIJKLMNOPQRSTUVWXYZ01'],

MEDIUMHard-coded secrets · secret.aws · CWE-798, CWE-321

tests/core/privacy-filter.test.ts:114

['aws access key id', 'AKIAABCDEFGHIJKLMNOP']